Comment on Federal judge sides with Meta in lawsuit over training AI models on copyrighted books
mindbleach@sh.itjust.works 1 week ago
Turning books into a language model is transformative. No LLM is a substitute for the original works.
Comment on Federal judge sides with Meta in lawsuit over training AI models on copyrighted books
mindbleach@sh.itjust.works 1 week ago
Turning books into a language model is transformative. No LLM is a substitute for the original works.
p03locke@lemmy.dbzer0.com 1 week ago
Yep. As much as everybody wants to shit on Zuckerberg, you can’t recreate exact copies with LLMs of any sort. You can’t claim that a 12GB image creation model somehow houses the entirety of all human-generated images.
KoboldCoterie@pawb.social 1 week ago
Really, the whole basis for the anti-AI arguments seem to boil down to “It feels wrong that a billionaire’s corporation should be able to take the work of artists and writers and, without paying them for it, use it to create a tool that is then used to put them out of work.” And that’s absolutely 100% true, but it unfortunately doesn’t hold any legal weight, and the terms we currently have to describe intellectual property theft simply aren’t sufficient to describe what’s going on here. Until new laws are passed, I don’t see any of these attempts to stop AI going anywhere, but I’d love to be proven incorrect.
mindbleach@sh.itjust.works 1 week ago
Most anti-AI sentiments seem like misplaced hatred of awful companies forcing nonsense on everybody, or a refusal to place judgement on capitalism itself.
mindbleach@sh.itjust.works 1 week ago
Right, a 12 GB model trained on 100,000,000 images isn’t big enough to contain an MD5 checksum of each.
The same people expect it to identify the authorship of sentence fragments, but never quote one whole paragraph from any book ever. Now: gigabytes of text could be a significant fraction of all books. But finding a single recognizable page is news. Storing text is not what these companies spent a bajillion dollars on.